State-Variable Models of Pharmacodynamic Tolerance

Why Represent Tolerance as a State?

A drug effect can weaken during repeated or sustained exposure even when the measured concentration remains similar. The operational description of tolerance is a reduced response under comparable exposure; a state-variable model adds a possible time course for that change. Its central idea is that the system carries forward information about prior exposure: a tolerance state builds while the drug drives adaptation and declines when that drive is small.

This is different from making the response depend only on the current concentration. Two observations at the same concentration can have different predicted effects if the tolerance state differs. The state is a compact model component, not by itself a claim that a particular receptor, mediator, or cellular process caused the response change. The earlier material on receptor desensitization, receptor internalization, and counter-regulatory mediators provides distinct biological interpretations that may motivate a state.

Definition: A tolerance state $T(t)$ is a time-varying model variable that summarizes an adaptive influence of past exposure on the pharmacodynamic response. In the model below, larger $T(t)$ represents greater tolerance. The state builds through an exposure-dependent input and declines through a loss process. It is not necessarily a directly measured quantity.

A State Equation for Accumulation and Recovery

Let $C(t)$ be drug concentration and let $g(C)$ be a nonnegative exposure-drive function with $g(0)=0$. A common choice is a saturating function, $g(C)=C/(K_T+C)$, where $K_T>0$ has concentration units. Consider the state equation

\frac{dT(t)}{dt}=k_{\mathrm{b}}g(C(t))-k_{\mathrm{l}}(C(t))T(t).

Here $T(t)$ is dimensionless, $k_{\mathrm{b}}$ has units of inverse time, and $k_{\mathrm{l}}(C)$ is a positive loss rate with units of inverse time. The exposure input increases the state; the loss term removes state in proportion to the amount already present. A simpler and often useful special case uses a constant loss rate $k_{\mathrm{l}}(C)=k_{\mathrm{r}}$. Allowing the loss rate to depend on exposure is more flexible, but it adds a modeling assumption that should be supported by data or biological rationale.

For a constant concentration over an interval, write $g=g(C)$ and $k_{\mathrm{l}}=k_{\mathrm{l}}(C)$. The state approaches the interval-specific equilibrium $T_{\mathrm{ss}}=k_{\mathrm{b}}g/k_{\mathrm{l}}$. If exposure is absent, then $g(0)=0$, so the state decays exponentially at the low-exposure loss rate. These statements assume the rates are constant within the interval and that the state changes continuously according to the specified equation.

Model result: If $C(t)$ is constant on an interval of length $\Delta$, with constant $g(C)$ and $k_{\mathrm{l}}(C)>0$, then $$ T(t+\Delta)=T_{\mathrm{ss}}+\{T(t)-T_{\mathrm{ss}}\}e^{-k_{\mathrm{l}}(C)\Delta}, \qquad T_{\mathrm{ss}}=\frac{k_{\mathrm{b}}g(C)}{k_{\mathrm{l}}(C)}. $$ When $C=0$, this reduces to $T(t+\Delta)=T(t)e^{-k_{\mathrm{l}}(0)\Delta}$.

To justify the result, let $T_{\mathrm{ss}}=k_{\mathrm{b}}g(C)/k_{\mathrm{l}}(C)$ on the constant-exposure interval. Subtracting this constant from both sides of the differential equation gives

\frac{d}{dt}\{T(t)-T_{\mathrm{ss}}\} =-k_{\mathrm{l}}(C)\{T(t)-T_{\mathrm{ss}}\}.

The solution of this first-order equation is $T(t+\Delta)-T_{\mathrm{ss}}=\{T(t)-T_{\mathrm{ss}}\}e^{-k_{\mathrm{l}}(C)\Delta}$, which gives the stated expression. Because the exponential factor lies between zero and one for $\Delta>0$, the state moves toward the interval-specific equilibrium without overshooting it under these assumptions.

Worked Example: Accumulation Followed by Low-Exposure Decay

Consider hypothetical exposure that strongly saturates the drive function from hour 0 through hour 6, so $g(C)=1$, followed by negligible exposure from hour 6 through hour 11, so $g(C)=0$. Let $T(0)=0$ and $k_{\mathrm{b}}=0.200\ \mathrm{h}^{-1}$. During the high-exposure interval, let $k_{\mathrm{l}}=0.300\ \mathrm{h}^{-1}$; during low exposure, let it be $0.100\ \mathrm{h}^{-1}$. This illustrative model allows different loss rates in the two exposure regimes.

During hours 0–6, the state equilibrium is $T_{\mathrm{ss}}=0.200/0.300=0.6667$. Therefore,

T(6)=\frac{0.200}{0.300}\left(1-e^{-0.300(6)}\right) =\frac{0.200}{0.300}(1-e^{-1.800}) =0.5564674.

During the next five hours, there is no input and the low-exposure loss rate is $0.100\ \mathrm{h}^{-1}$. Thus,

T(11)=T(6)e^{-0.100(5)} =0.5564674e^{-0.500} =0.3375135.

The state has fallen from 0.5565 to 0.3375, but it has not returned to zero. The calculation also shows why the loss rate must be specified for the relevant exposure regime: using $0.300\ \mathrm{h}^{-1}$ for the five-hour low-exposure interval would give a different result.

Connecting the State to Potency or Efficacy

Once a state equation is specified, it must be linked to the response model. A potency-linked formulation makes the concentration required for a given response increase with tolerance. For example, define a state-dependent $EC_{50}$ by $EC_{50}(T)=EC_{50,0}(1+\beta T)$, with $EC_{50,0}>0$ and $\beta\geq 0$. With a Hill coefficient $\gamma>0$, a response above baseline can then be written as

E(t)-E_0 = E_{\max} \frac{C(t)^\gamma} {\{EC_{50,0}(1+\beta T(t))\}^\gamma+C(t)^\gamma}.

Here $E_0$ is the baseline response, and $E_{\max}$ is the maximum response above baseline in this formulation. When $\beta>0$, increasing $T$ raises the state-dependent $EC_{50}$; at a fixed positive concentration, that lowers the modeled response. When $\beta=0$, the state has no effect on potency. An alternative is to make the maximum effect state-dependent, for example $E_{\max}(T)=E_{\max,0}/(1+\delta T)$, where $\delta\geq0$. These are alternative ways to represent tolerance, not interchangeable biological explanations. Which link is appropriate depends on the response pattern the model is intended to represent.

Worked Example: A Tolerance State That Reduces Apparent Potency

Use the state $T(11)=0.3375135$ from the preceding hypothetical exposure history. At hour 11, administer a short concentration challenge of $C=2.00\ \mathrm{mg/L}$; assume it is brief enough that the state does not materially change during the response measurement. Let $EC_{50,0}=2.00\ \mathrm{mg/L}$, $\beta=1.500$, $E_{\max}=10.00$ response units, $E_0=0$, and $\gamma=1$.

First calculate the state-dependent potency parameter:

EC_{50}(T)=2.00\{1+1.500(0.3375135)\} =2.00(1.50627025) =3.0125405\approx3.012541\ \mathrm{mg/L}.

The predicted response is then

E=10.00\frac{2.00}{3.012541+2.00} =\frac{20.00}{5.012541} =3.990\ \text{response units}.

At the same concentration with $T=0$, the model gives $E=10.00(2.00)/(2.00+2.00)=5.000$ response units. The model therefore predicts a lower response after the prior exposure history. This is a model-based comparison at the same concentration, not evidence by itself that a particular biological mechanism caused the difference.

Worked Example: An Alternative Link Through Maximum Effect

Keep $T=0.3375135$ and $C=2.00\ \mathrm{mg/L}$, but consider an efficacy-linked model with $E_{\max,0}=10.00$ response units, $\delta=0.800$, $EC_{50}=2.00\ \mathrm{mg/L}$, $E_0=0$, and $\gamma=1$. Here $T$ changes the maximum effect rather than $EC_{50}$.

The state-dependent maximum is

E_{\max}(T)=\frac{10.00}{1+0.800(0.3375135)} =\frac{10.00}{1.270011} =7.874\ \text{response units}.

Substituting into the concentration–response equation gives

E=7.874\frac{2.00}{2.00+2.00} =3.937\ \text{response units}.

This result is lower than the response predicted by the potency-linked example, but the two examples use different model links and different parameter values. Their numerical difference is not a basis for choosing between them. The relevant question is whether the available response data can support the assumed shape of the change.

Exposure Timing Can Matter Even at Similar Total Exposure

Because $T(t)$ retains a decaying influence of earlier exposure, total exposure alone generally does not determine the state at a particular time. The timing of exposure relative to the time of response measurement matters. This is one reason a state variable can describe a delayed loss of response that an instantaneous concentration–response relationship cannot.

Worked Example: Equal Exposure Durations, Different States

Compare two hypothetical schedules measured at hour 10. In both, exposure strongly saturates the drive ($g(C)=1$) for two hours, and is negligible otherwise. Use $k_{\mathrm{b}}=0.200\ \mathrm{h}^{-1}$, a high-exposure loss rate of $0.300\ \mathrm{h}^{-1}$, and a low-exposure loss rate of $0.100\ \mathrm{h}^{-1}$, with $T(0)=0$. Schedule A exposes from hour 0 to hour 2; Schedule B exposes from hour 8 to hour 10.

For either two-hour exposure interval, the state at its end is

T_{\mathrm{end}} =\frac{0.200}{0.300}(1-e^{-0.300(2)}) =\frac{0.200}{0.300}(1-e^{-0.600}) =0.300792.

For Schedule A, the state then decays for eight hours:

T_A(10)=0.300792e^{-0.100(8)} =0.13515.

For Schedule B, the exposure ends at hour 10, so $T_B(10)=0.300792$. The two schedules have the same exposure duration and, at the same concentration, the same exposure integral. Yet the modeled tolerance state at hour 10 is greater after the more recent exposure. This conclusion follows from the state dynamics and depends on the assumed exposure drive and loss rates.

Assumptions, Diagnostics, and Alternatives

The state equation is useful only insofar as its structure is defensible. Its basic assumptions are that the relevant exposure can be represented by $C(t)$, that the chosen drive function captures how exposure generates adaptation, and that loss is proportional to the current state at the specified rate. The model also assumes a single state is sufficient to summarize the history relevant to the response. None of these assumptions follows simply from observing a diminishing response.

Several patterns can indicate that the formulation is inadequate:

Predictions depend implausibly on the initial state. Check whether $T(0)$ is justified by the pre-study exposure history or needs to be estimated.
The fitted state becomes negative or grows without reasonable bounds. A negative state may contradict the intended meaning of tolerance; persistent growth may indicate a missing bound, an unsuitable drive function, or an incorrect exposure history.
Response residuals vary systematically with time or dose history. A single state, a single loss process, or the proposed potency/efficacy link may not capture the observed pattern.
State changes occur faster than the sampling schedule can resolve. Sparse observations may not distinguish rapid adaptation from an instantaneous concentration–response change.
Concentrations are measured inaccurately or incompletely. The state is driven by the supplied exposure history, so errors in that history can be mistaken for a pharmacodynamic adaptation pattern.

A direct concentration–response model is the simpler alternative when response depends only on current concentration and there is no evidence of history dependence. An effect-compartment model represents delayed equilibration between plasma concentration and an effect site; it does not, by itself, imply tolerance. A receptor or mediator model may give the changing response a more specific mechanistic structure, but requires corresponding biological assumptions and data. These approaches can sometimes produce similar response-time patterns, so the state-variable formulation should be viewed as a modeling choice rather than a unique explanation.

A common mistake is to interpret every decrease in response during dosing as proof that $EC_{50}$ has increased. The correction is to separate observation from model interpretation: a reduced response at comparable concentration is consistent with tolerance, but the data may also reflect changing exposure, delayed distribution, a changing maximum effect, or another time-varying process. Likewise, a state variable is not automatically a measured receptor fraction. Its scale and meaning are defined by its equations and parameterization.

Key takeaway: A tolerance state carries forward an exposure-dependent influence on response. Specify how exposure builds the state, how low exposure removes it, and how the state modifies the response model. Check the implied time course and avoid assigning a biological mechanism that the model and data do not establish.

Check Your Understanding

Answer each question, then compare your reasoning with the solutions.

In the state equation, what happens to $T(t)$ when $C=0$, and which parameter controls the decay timescale?
For constant exposure, explain what $T_{\mathrm{ss}}$ represents and why the state approaches it without overshooting under the stated model.
At fixed positive concentration, what does a positive $\beta$ do in the potency-linked model?
Why can two schedules with equal exposure integrals produce different tolerance states at the same measurement time?
Give one reason a delayed response should not automatically be interpreted as tolerance.

Solutions. (1) With $g(0)=0$, the equation becomes $dT/dt=-k_{\mathrm{l}}(0)T$, so $T$ decays exponentially; $1/k_{\mathrm{l}}(0)$ is its exponential time constant. (2) $T_{\mathrm{ss}}=k_{\mathrm{b}}g(C)/k_{\mathrm{l}}(C)$ is the equilibrium state under the constant exposure and rates. The solution is a weighted movement from the current state toward this value, with weight $e^{-k_{\mathrm{l}}\Delta}$ between zero and one. (3) Positive $\beta$ makes $EC_{50}(T)$ increase as $T$ increases, lowering the predicted response at fixed positive concentration when the other parameters are held fixed. (4) State is retained and then decays over time, so more recent exposure can leave a larger state at measurement even when total exposure is equal. (5) Delayed distribution to an effect site can also delay the response; delay alone does not establish a tolerance process.